Predicting Phenoconversion in Isolated RBD: Machine Learning and Explainable AI Approach

Area of Science:

• Neurology
• Sleep Medicine
• Computational Neuroscience

Background:

• Isolated rapid eye movement (REM) sleep behavior disorder (iRBD) is a prodromal marker for neurodegenerative diseases.
• Early prediction of phenoconversion timing and subtype is crucial for patient management.

Purpose of the Study:

• To develop and validate machine learning models for predicting phenoconversion timing in iRBD.
• To identify predictors for motor-first versus cognition-first neurodegenerative disease progression in iRBD patients.

Main Methods:

• Analysis of comprehensive clinical data from 178 iRBD individuals with a median follow-up of 3.6 years.
• Application of machine learning algorithms, including XGBSE-KN for timing and RandomForestClassifier for subtype prediction.
• Evaluation of model performance using concordance index, integrated Brier score, and Matthews correlation coefficient.

Main Results:

• The XGBSE-KN model accurately predicted phenoconversion timing (concordance index: 0.823).
• Factors associated with increased phenoconversion risk included older age, antidepressant use, and higher MDS-UPDRS Part III scores; coffee consumption showed a protective effect.
• The RandomForestClassifier distinguished between motor-first and cognition-first progression (MCC: 0.697), with higher MoCA scores and younger age predicting motor-first, and longer total sleep time predicting cognition-first outcomes.

Conclusions:

• Machine learning models offer valuable tools for predicting neurodegenerative disease development in iRBD patients.
• These predictive insights can facilitate tailored interventions and improve patient prognosis.
• Future research should incorporate additional biomarkers and external validation for broader applicability.